Django REST Framework custom fields validation

I will expand Konrad answer. I like it because its quite explicit, and also you are calling validation on other fields when we use them. So it is safer, probably will be redundant (some validations will be called twice)

First thing to note, is that if we implement like this, when we run run_validator, only validations set in validators variable will appear. So if we validate a field for example with the validate_ methods, it will not be run.

Also, I have make it inheritable, so we can reimplement the validation function and rehuse the code.

validators.py

from rest_framework.serializers import ValidationError

class OtherFieldValidator:

    #### This part is the same for all validators ####

    def __init__(self, other_field):
        self.other_field = other_field # name of parameter

    def set_context(self, serializer_field):
        self.serializer_field = serializer_field # name of field where validator is defined

    def make_validation(self,field, other_field):
        pass

    def __call__(self, value):
        field = value
        serializer = self.serializer_field.parent # serializer of model
        raw_other_field = serializer.initial_data[self.other_field] # data del otro campo

        try:
            other_field = serializer.fields[self.other_field].run_validation(raw_other_field)
        except ValidationError:
            return # if date_start is incorrect we will omit validating range

    #### Here is the only part that changes ####

        self.make_validation(field,other_field)

class EndDateValidator(OtherFieldValidator):

    def make_validation(self,field, other_field):
        date_end = field
        date_start = other_field
        if date_start and date_end and date_end < date_start:
            raise ValidationError('date cannot be')

So the serializer will be like this : serializers.py

# Other imports
from .validators import EndDateValidator

 def myfoo(value):                                                        
     raise ValidationError("start date error")                             

 class MyModelSerializer(serializers.ModelSerializer):                                        
     class Meta:                                                          
         model = MyModel                                                      
         fields = '__all__'                                                                                       
         extra_kwargs = {                                                 
             'date_end': {'validators': [EndDateValidator('date_start')]},
             'date_start': {'validators': [myfoo]},                       
         }                                                                

Another answer here might be useful, regarding the situation if one chooses to override serializer's validate() method.

Regarding answer on Order of Serializer Validation in Django REST Framework, I must say that serializer.validate() method is called at the end of the validation sequence. However, field's validators are called before that, in serializer.to_internal_value(), raising ValidationError at the end.

This means that custom validation errors do not stack with default ones.

In my opinion cleanest way to achieve desired behaviour is by using target field method validation in serializer class:

def validate_end_date(self, value):
    # validation process...
    return value

In case if you need another field value from the model, such as start_date in this case, you can get them (yet unvalidated, as a process is not complete) with:

# `None` here can be replaced with the field's default value
start_date = self.initial_data.get('start_date')

jgadelange's answer worked before django rest 3 probably. If any one using the django rest framework 3* version, I think this would be helpful for that folk. one should keep validation process in model level and clean method may be the one solution. But django rest framework announcement says here that, if someone wants to validate rest-call in model .clean method, he/she should override the serializer validate method and need to call the clean method form this serializer class by the following way

(because doc says : clean() method will not be called as part of serializer validation)

class MySerializer(serializers.ModelSerializer):

   def validate(self, attrs):
     instance = MyModel(**attrs)
     instance.clean()
     return attrs

and model

class MyModel(models.Model):
    start_date = models.DateField()
    end_date = models.DateField()

    def clean(self):
        if self.end_date < self.start_date:
            raise ValidationError("End date must be after start date.")

You should use an object wide validation (validate()), since validate_date will never be called since date is not a field on the serializer. From the documentation:

class MySerializer(serializers.ModelSerializer):
    def validate(self, data):
        """
        Check that the start is before the stop.
        """
        if data['start_date'] > data['end_date']:
            raise serializers.ValidationError("finish must occur after start")
        return data

As suggested by Michel Sabchuk you can add the validation error to the end_date field:

class MySerializer(serializers.ModelSerializer):
    def validate(self, data):
        """
        Check that the start is before the stop.
        """
        if data['start_date'] > data['end_date']:
            raise serializers.ValidationError({"end_date": "finish must occur after start"})
        return data

Another possibility is to create a validator. I created one based on the code for UniqueTogetherValidator:

from rest_framework.utils.representation import smart_repr

class DateBeforeValidator:
    """
    Validator for checking if a start date is before an end date field.
    Implementation based on `UniqueTogetherValidator` of Django Rest Framework.
    """
    message = _('{start_date_field} should be before {end_date_field}.')

    def __init__(self, start_date_field="start_date", end_date_field="end_date", message=None):
        self.start_date_field = start_date_field
        self.end_date_field = end_date_field
        self.message = message or self.message

    def __call__(self, attrs):
        if attrs[self.start_date_field] > attrs[self.end_date_field]:
            message = self.message.format(
                start_date_field=self.start_date_field,
                end_date_field=self.end_date_field,
            )
            # Replace the following line with
            #   raise serializers.ValidationError(
            #       {self.end_date_field: message},
            #       code='date_before',
            #   )
            # if you want to raise the error on the field level
            raise serializers.ValidationError(message, code='date_before')

    def __repr__(self):
        return '<%s(start_date_field=%s, end_date_field=%s)>' % (
            self.__class__.__name__,
            smart_repr(self.start_date_field),
            smart_repr(self.end_date_field)
        )


class MySerializer(serializers.ModelSerializer):
    class Meta:
        # If your start/end date fields have another name give them as kwargs tot the
        # validator:
        #   DateBeforeValidator(
        #       start_date_field="my_start_date", 
        #       end_date_field="my_end_date",
        #   )
        validators = [DateBeforeValidator()]

Pre DRF 3.0 you could also add it to the clean function of a model, but this is not called anymore in DRF 3.0.

class MyModel(models.Model):
    start_date = models.DateField()
    end_date = models.DateField()
    def clean(self):
        if self.end_date < self.start_date:
            raise ValidationError("End date must be after start date.")